An artificial intelligence-based system manages to design and run real chemistry experiments, showing its potential to help scientists make more discoveries faster. The system was named “Coscientist”.

Within minutes and on the first try the system was able to learn complex chemical reactions and design a successful laboratory procedure for their production.

The system uses large language models, including OpenAI’s GPT-4, to perform the full range of the experimental process with a simple language prompt. The system was able to do tasks that all research chemists do: look up information about chemical compounds, find and read technical manuals on how to control robotic lab equipment, write computer code to conduct experiments, and analyze the data that emerge to determine what worked and what didn’t.

The system, developed by researchers at Carnegie Mellon University, led by Assistant Professor of Chemistry and Chemical Engineering Gabe Gomes, is featured in a publication in the journal Nature.

In the publication the research team describes that the Coscientist can design the chemical synthesis of known compounds, search for, and use documentation to execute high-level commands in an automated laboratory. Also, test fluid handling instruments, complete scientific tasks that require the use of multiple computing units and different data sources, and solve optimization problems by analyzing previously collected data.

As the research team notes, “we expect that intelligent systems of autonomous scientific experimentation will bring about huge discoveries, unanticipated treatments and new materials. While we cannot predict what these breakthroughs will be, we hope to see a new way of conducting research arising from the collaboration between humans and machines.” The primary goal, they add, is to ensure the ethical and responsible use of these tools.

In early 2024 Carnegie Mellon University in collaboration with the “Emerald Cloud Lab”, a research facility founded by University alumni and operated remotely, will create a corresponding cloud lab. The Carnegie Mellon University Cloud Lab will give researchers access to more than 200 pieces of equipment, and the technologies presented in the paper will be used there.

*The scientific publication: https://www.nature.com/articles/s41586-023-06792-0